Hydrol. Earth Syst. Sci. Discuss., 4, 439–473, 2007
www.hydrol-earth-syst-sci-discuss.net/4/439/2007/
© Author(s) 2007. This work is licensed
under a Creative Commons License.
Hydrology and
Earth System
Sciences
Discussions
Papers published in Hydrology and Earth System Sciences Discussions are under
open-access review for the journal Hydrology and Earth System Sciences
Updated world map of the Köppen-Geiger
climate classification
1
2
M. C. Peel , B. L. Finlayson , and T. A. McMahon
1
1
Department of Civil and Environmental Engineering, The University of Melbourne, Victoria,
Australia
2
School of Anthropology, Geography and Environmental Studies, The University of Melbourne,
Victoria, Australia
Received: 15 February 2007 – Accepted: 16 February 2007 – Published: 1 March 2007
Correspondence to: M. C. Peel ([email protected])
439
Abstract
5
10
15
Although now over 100 years old, the classification of climate originally formulated
by Wladimir Köppen and modified by his collaborators and successors, is still in
widespread use. It is widely used in teaching school and undergraduate courses on
climate. It is also still in regular use by researchers across a range of disciplines as a
basis for climatic regionalisation of variables and for assessing the output of global climate models. Here we have produced a new global map of climate using the KöppenGeiger system based on a large global data set of long-term monthly precipitation
and temperature station time series. Climatic variables used in the Köppen-Geiger
system were calculated at each station and interpolated between stations using a
two-dimensional (latitude and longitude) thin-plate spline with tension onto a 0.1◦ ×0.1◦
grid for each continent. We discuss some problems in dealing with sites that are not
uniquely classified into one climate type by the Köppen-Geiger system and assess
the outcomes on a continent by continent basis. Globally the most common climate
type by land area is BWh (14.2%, Hot desert) followed by Aw (11.5%, Tropical savannah). The updated world Köppen-Geiger climate map is freely available electronically
at http://www.hydrol-earth-syst-sci.net/????.
1 Introduction
20
25
The climate classification based on the work of Wladimir Köppen, and dating from
1900, continues to be the most widely used climate classification over a century later.
Essenwanger (2001) has provided a comprehensive review of the classification of climate from prior to Köppen through to the present. The period of most activity was
from the mid-nineteenth century through to the 1950s. What is somewhat surprising
about this time profile of activity is that as both data availability and computing power
to process it has become increasingly widely available post-1960, the level of activity in
the development of new climate classifications has markedly declined. The continued
440
5
10
15
20
25
popularity and widespread use of the Köppen classification is remarkable. There is no
doubt an element of historical inertia in this as each generation of students is taught
global climate using this system and it is the basis of most common global climate
maps. To replace it with a new system would be a significant task. Arthur Wilcock
(1968) was probably correct in surmising: “If . . . . . . . one is convinced that there are
in principle strict limits to what can be achieved by any simple classification, one may
consider it profitless to seek minor improvements at the cost of confusion.” (p13).
Köppen’s inspiration for developing a world map of climate classification in 1900
owed much to the global vegetation map of Grisebach published in 1866 and Köppen’s
own background in plant sciences (Wilcock, 1968). Thornthwaite (1943) claims that
Köppen’s use of the first five letters of the alphabet to label his climate zones is taken
from the five vegetation groups delineated by the late nineteenth century French/Swiss
botanist Alphonse De Candolle who in turn based these on the climate zones of the
ancient Greeks. It is inconceivable that Köppen could have produced his original classification and map without using other landscape signals of climate (particularly vegetation) since there would have been so little observed climate data available at that
time. In Figure 3 of this paper we show the relative number of stations with temperature
and precipitation data starting from 1800. Compared with what is available now, there
would have been data from few stations available to Köppen and the global distribution
would have been much more inconsistent than is the case now. In the light of this, the
persistence of his scheme of classification is even more remarkable.
While Sanderson (1999) has argued that it is time for modern atmospheric scientists
to develop a new classification of climates, the Köppen classification continues to be
the one most widely used in teaching. If we take as an example the textbooks of
Arthur Strahler that are in very wide use in the English speaking world, it is the case
that despite Strahler’s own attempt to produce a new climate classification (see, for
example, Strahler, 1971) the latest edition of this series of texts still uses the Köppen
system (Strahler and Strahler, 2005).
The use of Köppen’s classification is not confined to teaching. Many researchers
441
5
10
15
20
25
routinely use it for their own particular research purposes. The present authors have
used it as the basis for grouping rivers into like climate areas around the world in order
to facilitate comparisons of runoff characteristics (McMahon et al., 1992; Peel et al.,
2004). Lohmann et al. (1993) have applied the Köppen classification to the output of
both atmosphere general circulation models and coupled atmosphere-ocean circulation models and compared these to maps of the Köppen classification using modern
data sets and to Köppen’s 1923 map. They modelled both present conditions and enhanced greenhouse scenarios and concluded: “However, the Köppen classification is
easier to apply and is still a useful tool for estimating the ability of climate models to
reproduce the present climate as well as indicate the impact of climate changes on the
biosphere.” (p 191) No doubt Köppen would have been pleased with this assessment.
In a similar study to that of Lohmann et al. (1993), Kalvova et al. (2003) compared
global climate model outputs to maps of Köppen’s classification drawn from gridded
observed data and to Köppen’s 1923 map. They were attracted to the Köppen system
because of its known links to natural vegetation patterns as they have attempted to
assess the impact of global warming on major biomes. They also compare the map
they produced of Köppen’s climate zones based on modern data with his 1923 map
and show that the differences are only around 0.5% of the area distribution. Similar
uses of the Köppen classification have been made by Wang and Overland (2004),
Gnanadesikan and Stouffer (2006) and Kleidon et al. (2000) where it is the relation
between the Köppen zones and natural vegetation systems that has made it useful to
their purposes. It is noteworthy that Kleidon et al. (2000) also used the Köppen 1923
map as a basis for comparison.
A more critical approach to the Köppen classification has been taken by Triantafyllou and Tsonis (1994) who claim to be the first to evaluate the Köppen classification
using modern temperature and precipitation data (for the northern hemisphere). They
classified climate stations on a year by year basis and then analysed the frequency
with which they shifted between the major Köppen climate types (e.g. A to B) in order
to assess the adequacy of the Köppen system. In North American and North Africa
442
5
10
15
20
25
they found low variability within a climate type and narrow regions of high variability
between climate types, indicating the Köppen system performed adequately. For Europe and Asia they found the pattern of variability less defined, indicating either high
within climate type variability or wide regions between climate types resulting in an inadequate performance of the Köppen system. It is the case though that the Köppen
classification was intended to represent long term mean climate conditions and not
year-to-year variability though it can be put to good use as the basis for assessing climate variability on a year-to-year (Dick, 1964) or multi-decadal basis (Gentilli, 1971).
Triantafyllou and Tsonis (1994) conclude, with Sanderson (1999), that there is a need
for a new scheme to represent the world’s climates.
That may be so, but when Fovell and Fovell (1993) used cluster analysis to objectively determine climate zones for the conterminous United States based on modern
climate data they returned to the Köppen classification to assess the outcomes. Similarly Stern et al. (2000), with all the data resources of the Australian Bureau of Meteorology at their disposal, used a modification of the Köppen classification to draw a new
map of the climates of Australia. Their assessment that “. . . the telling evidence that
the Köppen classification’s merits outweigh its deficiencies lies in its wide acceptance.”
(p 2).
It is against this background that we have chosen to re-draw the Köppen-Geiger
world map using global long-term monthly precipitation and temperature station data.
Recently, three Köppen world maps based on gridded data have been produced for
various resolutions, periods and levels of complexity. Kalvova et al. (2003) using Climate Research Unit (CRU, the University of East Anglia) gridded data for the period
1961–1990 presented a map of the 5 major Köppen climate types (with E divided into
◦
◦
2 types) at a resolution of 2.5 latitude by 2.5 longitude. Gnanadesikan and Stouffer
(2006) presented a Köppen map of 14 climate types based on the same CRU data and
period as Kalvova et al. (2003), but at a resolution of 0.5◦ latitude by 0.5◦ longitude.
The most comprehensive Köppen world map drawn from gridded data to date is that of
Kottek et al. (2006) who presented a map with 31 climate types at a resolution of 0.5◦
443
◦
5
10
latitude by 0.5 longitude based on both the CRU and Global Precipitation Climatology
Centre (GPCC) VASClimO v1.1 data sets for the period 1951–2000.
All three maps based on gridded data are for restricted periods (1961–1990 or 1951–
2000) and any sub-grid resolution climate type variability has been obscured. So here
we present an updated world map of the Köppen-Geiger climate classification based
on station data for the whole period of record. The data and methodology used to
construct this map are described in the next section. Individual continental KöppenGeiger climate maps are presented and discussed. The continental maps are then
combined to form the new world Köppen-Geiger map, which is followed by a discussion,
a link to the map for free download and a conclusion.
2 Data and methodology
15
20
25
The philosophy behind the construction of this updated version of the Köppen-Geiger
climate map is to rely on observed data, rather than experience, wherever possible and
minimise the number of subjective decisions. To this end, a large, globally extensive,
climatic dataset was used to describe the observed climate and the methodology used
to interpolate the observations was chosen to be simple and flexible, but not beyond
what the data could support. There have been many modifications proposed to the
Köppen system but here we have used criteria that follow Köppen’s last publication
about his classification system in the Köppen-Geiger Handbook (Köppen, 1936), with
the exception of the boundary between the temperate (C) and cold (D) climates. We
◦
have followed Russell (1931) and used the temperature of the coldest month >0 C,
◦
rather than >–3 C as used by Köppen in defining the temperate – cold climate boundary (see Wilcox, 1968 and Essenwanger, 2001 for a history of this modification).
Long-term station records of monthly precipitation and monthly temperature were
obtained from the Global Historical Climatology Network (GHCN) version 2.0 dataset
(Peterson and Vose, 1997). Stations from this dataset with at least 30 observations
for each month were used in the analysis (12 396 precipitation and 4844 tempera444
5
10
15
20
25
ture stations). Figures 1 and 2 show the global spatial distribution of precipitation and
temperature stations respectively. In Fig. 2 temperature stations that also have precipitation data are denoted separately from the temperature only stations. Regions of high
station density are the USA, southern Canada, northeast Brazil (precipitation only), Europe, India (precipitation only), Japan and eastern Australia. Desert, polar and some
tropical regions, like Saharan Africa, Saudi Arabia, central Australia, northern Canada,
northern Russia and the Amazon region of Brazil have sparse station density.
In the following analysis the complete period of record at each precipitation and temperature station is used. The stations exhibit a wide range of record lengths from a
minimum of 30 values for each month up to 299 for precipitation and 297 for temperature. In Fig. 3, the percentage of stations with a value for a given month is plotted over
time and shows that the historical period that the data are most representative of are
from 1909 to 1991 for precipitation and 1923 to 1993 for temperature. Spatially there
is variation in the period of record covered, with Australia, Europe, Japan and the USA
generally having the longest records.
The whole of record approach assumes that data from one period is comparable with
data from any other period. This assumption can be violated by global or local trends,
like the recent observed warming of global surface temperature, largely attributable to
increasing concentrations of greenhouse gases (Barnett et al., 2005). However, at the
level of broad climate types (1st letter, see Table 1) the Köppen-Geiger climate classification has been found to be relatively insensitive to temperature trends (Triantafyllou
and Tsonis, 1994). The greatest sensitivity of the resultant map to climatic trends is
likely to be in the transition zones between climate types, rather than within a climate
type. Any benefits of restricting the data to a common period would be achieved at the
expense of spatial representativeness, since stations with data that did not sufficiently
overlap the common period would have been excluded from the analysis. In using the
complete period of record the resultant Köppen-Geiger climate type map presents the
long-term climate type for the maximum number of locations around the world.
A total of 4279 locations have data for both precipitation and temperature (indicated
445
5
10
15
20
25
with an “x” in Fig. 2). At these locations the Köppen-Geiger climate type can be determined from the raw station data. Although an updated Köppen-Geiger map based
on 4279 locations would be a significant improvement over earlier maps, considerable information would be lost if locations with only precipitation or temperature were
ignored. To avoid this loss of information the precipitation and temperature based variables used in the criteria to determine the Köppen-Geiger climate type (see Table 1)
were calculated at each precipitation and temperature station. Then each variable was
interpolated using a two-dimensional (latitude and longitude) thin-plate spline with tension (Mitas and Mitasova, 1988) onto a 0.1×0.1 degree of latitude and longitude grid
for each continent. The Köppen-Geiger criteria were then applied to the splined variables. The tension spline interpolations were performed in ESRI ArcMap version 9.1
using settings of “weight” = 1 and “points” = 10 for all interpolations. The chosen tension spline settings provided flexible and smooth interpolations of the climatic variables
across the wide range of station densities and variable values experienced within and
between continents. Although optimisation of the spline settings for each variable and
continent may have improved the interpolation of individual climatic variables, it was
not expected to significantly alter the final Köppen-Geiger climate type maps (which is
a combination of the variables) and so was not pursued in this paper. However, in continental sub-regions with high station density more flexible spline settings would have
provided better local results.
A potential improvement to the above methodology would be to apply a threedimensional spline, using elevation as the third dimension (based on a digital elevation
model like HYDRO1k, USGS, 2000). Elevation plays a key role in the observed spatial
pattern of precipitation and temperature and is an important variable in the interpolation
of precipitation and temperature fields (Daly, 2006). In this analysis long-term average
climatic variables are interpolated, which display considerably less spatial variability
than is typically observed in daily, monthly or even annual fields to which 3-D interpolation is usually applied. However, due to sparse station density, particularly in high
elevation regions, a more complex 3-D interpolation procedure was not used in this
446
5
10
15
20
25
paper.
Another potential improvement would be to conduct cross-validation on the individual
splined variables to investigate the sensitivity of each spline to the underlying data as
well as cross-validation of the final Köppen-Geiger climate type map to assess the
sensitivity of the map to each individual station. Such a sensitivity analysis would likely
indicate that the Köppen-Geiger climate type map is most sensitive in regions of low
station density, transition zones between climate types and also in regions where there
are large elevation differences. A cross-validation exercise was not conducted as part
of this paper due to the considerable amount of time required for such an exercise.
A description of the symbols and the criteria used to define the Köppen-Geiger climate types is provided in Table 1. The 30 possible climate types in Table 1 are divided
into 3 tropical (Af, Am and Aw), 4 arid (BWh, BWk, BSh and BSk), 9 temperate (Csa,
Csb, Csc, Cfa, Cfb, Cfc, Cwa, Cwb and Cwc), 12 cold (Dsa, Dsb, Dsc, Dsd, Dfa, Dfb,
Dfc, Dfd, Dwa, Dwb, Dwc and Dwd) and 2 polar (ET and EF). All precipitation variables
are in units of millimetres (mm) and all temperature variables are in units of degrees
Celsius (◦ C). Since all locations that satisfy the B climate criteria will also satisfy one
of the other (A, C, D or E) climate criteria, the B climates must be identified first. The
set of locations defined as having a B climate is based on a combination of mean annual precipitation and mean annual temperature. The sets of A, C, D and E climate
locations are mutually exclusive and are based on temperature criteria only.
In applying the criteria to the global dataset it became apparent that in some C
climate locations it is possible to satisfy both the Cs and Cw criteria simultaneously.
An example of this is Herberton Post Office (17.38 S, 145.38 E, elevation 900 m) in
North Queensland, Australia. Table 2 shows the monthly averages of precipitation and
temperature for the Herberton Post Office record. Each average is based on about
105 monthly values for precipitation and 78 monthly values for temperature, so these
averages are from a long-term station. Following the Köppen-Geiger criteria in Table 1,
summer is the period ONDJFM and over 70% of the mean annual precipitation falls
during summer. Since the mean annual precipitation (1146 mm) is larger than the B
447
5
10
15
20
25
climate type precipitation threshold of 678 mm (10 * (2 * 19.9 + 28)) (see Table 1) the
station is not a B climate. With the coldest monthly temperature 15.5◦ C, the station
is not an A climate, but does satisfy the criteria for a C climate. When the second
letter is allocated, the driest summer month is 25.3 mm, which is below 40 mm and is
less than one third of the wettest winter month (86.0 mm), so the station satisfies a Cs
climate type. However, the driest winter month (16.5 mm) is less than one tenth of the
wettest summer month (238.4 mm), so this station also satisfies the Cw climate type.
Considering that over 70% of the precipitation falls during summer a classification of
Cw is appropriate, while a classification of Cs is not.
For cases such as Herberton Post Office an additional test was developed and applied to determine whether to use Cs or Cw at a given location. When a location satisfied both Cs and Cw criteria, the precipitation for the six months containing summer
and the six months containing winter were compared, with Cw assigned if the summer
precipitation was greater than the winter precipitation. This additional rule was also
applied to locations that satisfied both the Ds and Dw climate criteria as well.
In the presence of steep gradients of a variable, splines are known to over- or underestimate, particularly if there is limited station density. Initially, these spline artefacts
were ignored at the individual variable level. However, once a continental map of
Köppen-Geiger climate type was produced, the map was checked for locations where
over or under-estimation in a splined variable had caused an aberrant climate type to
be defined. These aberrant areas were generally found to be small and were patched
by hand based on surrounding observations. Conversely, in some low station density
regions, a climate type would extend further than expected due to a lack of observations to inform the spline. In these situations, discussed later, the climate type was left
unmodified, due to a lack of information on which to base any corrections.
Contrary to some earlier forms of the Köppen-Geiger climate type map the concept
of a high altitude climate type is not used in this paper. Designated as a separate 1st
letter climate type H (Highland) or as a sub category of the E climate type (ETH or
EFH) the highland climate type requires elevation information to be defined. Although
448
5
10
global elevation data are available, defining all locations above a specified elevation as
an H climate provides little information about the climate at those locations relative to
using the full suite of A, B, C, D and E climate types. Therefore no highland climate
type is used in this paper.
Once the Köppen-Geiger climate type map for each continent was constructed, the
percentage of land area covered by the major climate types was calculated. Since the
area of a 0.1×0.1 degree pixel changes with latitude, a map of 0.1×0.1 degree pixel
area was constructed and then projected onto a Cylindrical Equal Area projection of
2
the world to determine the area (in km ) of each 0.1×0.1 degree pixel. These pixel
areas were then summed for each climate type to provide an estimate of the land area
covered by each climate type.
3 Continental maps
15
The Köppen-Geiger climate type map was determined using the methodology described in the previous section for each continent. These maps are presented and
discussed in this section.
3.1
20
25
Africa
The map of Köppen-Geiger climate type for Africa (Fig. 4) shows that only three (A, B
and C) of the main climate types are present in Africa. Of these three the dominant
climate type by land area is the arid B (57.2%), followed by tropical A (31.0%) and
temperate C (11.8%).
Figure 4 is based on 1436 precipitation and 331 temperature stations. Of these stations a total of 313 had both precipitation and temperature data for the same location,
from which the climate type could be independently calculated and checked against
the map. The climate type at 309 of these locations matched the map exactly and in
the remaining 4 locations the correct climate type was present in a neighbouring cell.
449
5
10
15
20
25
The thin-plate spline with tension settings used to interpolate individual variables
does not force the spline through the exact station values. Therefore, slight differences
in climate type between station and spline-based estimates are expected. If the station
and map climate types do not match exactly, then the station-based climate type usually
appears in a neighbouring cell of the spline-based map. However, if the station-based
climate type is not present in a neighbouring cell, then the spline-based estimate is
incorrect, indicating that at least one of the splines has not successfully interpolated
the observed data. For Africa, all of the station estimates match the map exactly or are
in a neighbouring cell.
The low density of temperature stations in Africa resulted in some climate types
extending further than expected, which could not be corrected due to lack of data.
The two regions where this is most evident are the temperate regions in the Eastern
Rift Valley south of Nairobi, Kenya and around Antananarivo, Madagascar. In both
of these cases the temperature stations are in high elevation locations (Nairobi and
Antananarivo) and experience a temperate climate type. However, due to the lack
of nearby lower elevation temperature stations, the temperate influence of both these
high elevation stations extends well beyond their immediate location and large regions
of temperate climate type result in regions that are more likely to be tropical.
Another interesting feature in Fig. 4 is the division of the temperate region in the
Ethiopian highlands into Cs and Cw climate types. This region experiences a strong
seasonal precipitation regime (most precipitation occurs during April–September) but
◦
a very even monthly temperature regime (the hottest month is ∼5 C hotter than the
coldest month). On the western side of the highlands, the six months from October–
March are warmer than April–September, so summer is defined as October–March,
thus the precipitation falls in winter. Whereas on the eastern side of the highlands
the six months from April–September are warmer, so summer is defined as April–
September, thus the precipitation falls in summer. Therefore, in Fig. 4 the western side
of the highlands have a Cs climate and the eastern side have a Cw climate.
450
3.2
5
10
15
20
25
Asia
In this paper Asia is defined as the region east of a north-south line through the Urals
Mountains down to the Arabian Sea. The map of Köppen-Geiger climate type for Asia
(Fig. 5) shows that all five of the main climate types are present in Asia. The dominant
climate type by land area is the cold D (43.8%), followed by arid B (23.9%), tropical A
(16.3%), temperate C (12.3%) and polar E (3.8%).
Figure 5 is based on 3650 precipitation and 944 temperature stations. Of these stations a total of 748 had both precipitation and temperature data for the same location,
from which the climate type could be independently calculated and checked against
the map. The climate type at 732 of these locations matched the map exactly and in
14 of the remaining 16 locations the correct climate type was present in a neighbouring
cell. However, at 2 locations (Kodaikanal, India and Pune, India) the observed climate
type differs from the map, indicating that at least one of the splines has not successfully
interpolated the observed data.
In the case of Kodaikanal (10.2 N, 77.5 E), located at 2343 m elevation on the eastern
side of the Western Ghats in Tamil Nadu State, the seasonality of precipitation is much
less distinct than that observed in the surrounding lowlands (∼100 m elevation). So
although the observations indicate a Cfb climate type at Kodaikanal, the map indicates
a Cwb climate type. The disparity is due to the high density of precipitation stations
in the surrounding lowlands, which influence the precipitation splines, particularly the
winter dry month spline, to underestimate the precipitation at Kodaikanal. This underestimation, particularly the winter dry month precipitation, leads to a Cw rather than
a Cf climate type. The universal spline settings used for the analysis are not flexible
enough to fully capture the local variation in precipitation, resulting in the misallocation of this site. Correcting the final climate type map around Kodaikanal would be
speculative and has not been attempted.
At Pune (18.5 N, 73.9 E), in Maharashtra State, there is a very strong orographic gradient in mean annual precipitation. The thin coastal strip to the west (25 km) of Pune
451
5
10
15
receives mean annual precipitation of between 3000–6000 mm. At Pune the mean
annual precipitation is 700 mm. The observations at Pune indicate a BSh climate,
whereas the map indicates an Aw climate. This difference is due to the mean annual
precipitation spline overestimating the precipitation at Pune, due to the very steep precipitation gradient, and thus placing Pune in an A rather than a B climate type. Again,
the universal spline settings used are not flexible enough to fully capture the local variation in precipitation, resulting in the misallocation of this site. Likewise, correcting the
final climate type map around Pune would be speculative and has not been attempted.
The low density of temperature stations in southern India resulted in a region of
temperate climate type extending further south than expected. As mentioned above,
Kodaikanal in Tamil Nadu is a high elevation station in a temperate climate type. However, it is up to 140 km from its nearest neighbour to the west and over 400 km away
from the nearest station to the northeast. All the surrounding temperature stations are
in a tropical climate type. The large area of temperate, rather than tropical climate in
southern India is due to this single station and the lack of temperature stations in the
surrounding lowland areas, which could not be corrected due to lack of data.
3.3
20
25
North America
North America is defined in this paper as Canada, the USA, the countries of Central
America and the Caribbean Islands. The map of Köppen-Geiger climate type for North
America (Fig. 6) shows that all five of the main climate types are present in North
America. The dominant climate type by land area is cold D (54.5%), followed by arid B
(15.3%), temperate C (13.4%), polar E (11.0%) and tropical A (5.9%).
Figure 6 is based on 3034 precipitation and 2236 temperature stations. Of these stations a total of 2078 had both precipitation and temperature data for the same location,
from which the climate type could be independently calculated and checked against
the map. The climate type at 2020 of these locations matched the map exactly and in
all the remaining 58 locations the correct climate type was present in a neighbouring
cell.
452
5
10
15
20
25
In Guatemala, there is a low density of temperature stations, which results in a region of temperate climate type extending further than expected around Guatemala City
(14.5 N, –90.6 W). At an elevation of 1500 m, Guatemala City has a temperate Cwb
climate. However, it is up to 175 km from its nearest neighbour to the southeast in El
Salvador. All the surrounding temperature stations are in tropical climate types. Although much of the area around Guatemala City is likely to be temperate due to the
high elevation, the temperate region on the map extends into what are likely to be
tropical areas near the coast where there are no temperature stations to define the
boundary more closely.
A region of BWh south of Oaxaca, Mexico (17.1 N, –96.7 W) may be an artefact of the
spline process, but since there are no precipitation or temperature stations in this arid
region it is impossible to be certain. At an elevation of 1550 m, Oaxaca experiences an
arid BSh climate type, but whether the region to the south is as arid as BWh remains
uncertain.
In the methodology section, an additional criterion to deal with locations that satisfy
both Cs and Cw or Ds and Dw criteria was described. In North America this criterion
was successfully applied at 11 temperate locations (mainly in Mexico) with observed
precipitation and temperature data. However, other locations in North America show
that the criteria for selecting the second letter of the C and D climate types are not
always providing an appropriate outcome. For example, at Batopilas in the state of
Chihuahua, Mexico (27.0 N, 107.7 W), where only precipitation data are available, the
climate type in Fig. 6 is given as Csa. Nearby temperature stations confirm that summer
is in the six-month period from April through to September. At Batopilas the average
summer rainfall is 441 mm and the average winter rainfall is 191 mm. Nearby temperate
locations are all Cw, which is consistent with the summer dominant rainfall at Batopilas.
Following the temperate second letter criteria in Table 1, the driest month in summer
(5.3 mm) is below 40 mm and is less than one third of the wettest month in winter
(49.3 mm), so a Cs climate type is satisfied. When the Cw climate criterion is applied,
the driest month in winter (18.4 mm) is not one tenth of the wettest month in summer
453
5
10
15
(154.6 mm), so a Cw climate type is not satisfied. Since this site only satisfies the Cs
criteria, the additional check of the summer and winter precipitation, described in the
methodology section, is not applied. So at Batopilas, the summer rainfall is greater
than the winter rainfall, by a factor of 2, yet it is defined by the Köppen-Geiger criteria
as a temperate climate with a dry summer (Csa).
A slightly different concern with the temperate second letter criteria is observed at
White River, in the state of Arizona, USA (33.84 N, 109.97 W) where the KöppenGeiger criteria suggest a Csa climate type. However, a Cs climate appears inappropriate since summer precipitation (228 mm) is greater than the winter precipitation
(201 mm). A more appropriate climate type would be Cfa since the precipitation is
evenly distributed throughout the two six month periods. Other locations where concerns about the second letter criteria for the C or D climate types were observed
are Prescott (Cs –34.57 N, 112.44 W), USA and the Ds regions in northern Canada
and Alaska centred on Fort Yukon (66.6 N, 145.3 W), Anchorage (61.2, 149.9 W), Yellowknife (62.47 N, 114.45 W), Carcross (60.18 N, 134.7 W), Mayo (63.62 N, 135.87 W)
and west of Fairbanks (64.9 N, 147.7 W).
3.4
20
25
South America
The map of Köppen-Geiger climate type for South America (Fig. 7) shows that three
main climate types A, B and C dominate the climate of South America. Of these three
the dominant climate type by land area is tropical A (60.1%), followed by temperate C
(24.1%) and arid B (15.0%). The Polar E (0.8%) climate type occurs in four places in
South America, twice in the Andes, due to the high elevation, along the southern edge
of Chile (Strait of Magellan) / Argentina (Tierra del Fuego) and in the Falkland Islands
and South Georgia.
Figure 7 is based on 1115 precipitation and 192 temperature stations. Of these stations a total of 171 had both precipitation and temperature data for the same location
for which the climate type is known and can be checked against the map. The climate
type at 165 of these locations matched the map exactly and in 5 of the remaining 6
454
5
10
15
20
25
locations the correct climate type was present in a neighbouring cell. In the case of
Aracaju (10.9 S, 37.1 W), in the state of Sergipe, Brazil, the observed precipitation and
temperature satisfy the Am climate type (borderline Aw), whereas the map shows this
site in a region of Aw climate type. The splined variables indicate that this station is
clearly Aw, which is not consistent with the observations. The reason the splined variables are inconsistent with the observations is that this region has a very high density
of precipitation stations (Fig. 1). The universal spline settings used for the analysis are
not flexible enough to fully capture the local variation in precipitation, resulting in the
misallocation of this site.
The misallocation of Aracaju due to the spline settings not being flexible enough to
adequately represent the high density of precipitation stations is indicative of neighbouring parts of the Nordeste region of Brazil. The climate types represented in this
region BWh, BSh, Aw, Am and Af are consistent with the observations. However, the
spatial distribution of these climate types would be more in line with the precipitation
observations had a more flexible spline setting been used for the precipitation data in
this region. The sparse distribution of temperature stations in this region would make
any corrections to the final climate type map speculative.
The generally low density of temperature stations in South America resulted in some
climate types extending further than expected, which could not be corrected due to
the lack of data. Two regions where this is most evident are the temperate regions
northeast of Bogata, Colombia and around Quito, Ecuador. In both of these cases
the temperature stations are in high elevation locations and experience a temperate
climate type. However, due to the lack of nearby lower elevation temperature stations,
the temperate influence of both these high elevation stations extends well beyond their
immediate location and large regions of temperate climate type result in regions that
are more likely to be tropical. In the case of Bogata, a region of ET climate type is
observed within the temperate region. This region is largely coincident with the higher
sections of the Cordillera Oriental, which is likely to have a climate type of ET.
455
3.5
5
10
15
20
25
Europe
In this paper, Europe is defined as the region west of a north-south line through the
Urals Mountains down to the Arabian Sea and includes the Arabian Peninsula and the
countries of the Middle East. The Köppen-Geiger climate type map of Europe (Fig. 8)
shows that only four main climate types are found in Europe. The dominant climate
type by land area is cold D (44.4%), followed by arid B (36.3%), temperate C (17.0%)
and polar E (2.3%).
Figure 8 is based on 1209 precipitation and 684 temperature stations. Of these
stations a total of 496 had both precipitation and temperature data for the same location
for which the climate type is known and can be checked against the map. The climate
type at 488 of these locations matched the map exactly and in the remaining 8 locations
the correct climate type was present in a neighbouring cell.
As is the case for other continents, in Europe there are some climate types extending further than expected due to a low density of temperature stations. For example, around Mussala (42.18 N, 23.58 E), Bulgaria there is a region of ET climate type.
There are three temperature stations within this ET region, which are all located at
over 2000 m in elevation, and their observations confirm an ET climate type. However,
rather than being part of a single range of mountains, these three stations are located
on top of three separate peaks with extensive lowlands in between. As there are no
temperature stations located in the lowlands the splines ignore them and the entire
region is set to ET, rather than a mixture of ET and Df. A similar situation occurs in
the Central Massif of France around Le Puy de Dome (45.8 N, 2.9 E) and Mont Aigoual
(44.12 N, 3.58 E) where two temperature stations atop isolated peaks extend a region
of Df climate between them. This region should be a mixture of Df (on the mountains)
and Cf in the lowlands, but there are no temperature stations in the lowlands.
In Iceland the temperature stations are all located in coastal areas. Without any
inland temperature data to inform the splines, the coastal climate type of Dfc extends
inland where a climate type of ET is more likely to be appropriate. The region of ET
456
5
climate type in southern Norway extends to the coast at the northwestern edge, an
area that would be expected to be Cfb if temperature data were available.
Along the French-Italian border the low density of temperature stations effectively
reduces the area of ET climate type. At St Bernard (45.7 N, 6.9 E) in Switzerland the
climate type is ET, while the nearest station in a direct line to the south, Nice (43.65 N,
7.2 E) in France, experiences a Csa climate. Despite the Alps stretching almost to
the coast, the region of ET climate mapped does not, due to the lack of temperature
stations in the region.
3.6
10
15
20
The Köppen-Geiger climate type map of Australia (Fig. 9) shows that only three main
climate types are found in Australia. The dominant climate type by land area is arid B
(77.8%), followed by temperate C (13.9%) and tropical A (8.3%).
Figure 9 is based on 1807 precipitation and 351 temperature stations. Of these
stations a total of 345 had both precipitation and temperature data for the same location
for which the climate type is known and can be checked against the map. The climate
type at 340 of these locations matched the map exactly and in the remaining 5 locations
the correct climate type was present in a neighbouring cell.
The spatial distribution of precipitation and temperature stations in the arid western
central region of Australia is sparse, so boundaries between the different arid climate
types are less reliable than in the more data rich eastern and south western regions
(Fig. 2).
3.7
25
Australia
Islands, Greenland and Antarctica
In the spline analyses used to estimate the Köppen-Geiger climate type for each continent, data from small offshore islands were generally removed prior to analysis, since
they tended to influence the splines over the mainland in low station density regions.
Therefore, islands like the Canary Islands, Cape Verde, Hawaii and the islands of the
457
5
10
15
South Pacific were allocated a Köppen-Geiger climate type by hand based on observations from the islands.
In Greenland there were 10 precipitation and 9 temperature stations, 8 of which have
both variables at the same location. All of these stations are in coastal locations, so
there was no climatic information about the interior of Greenland on which to base a
spline analysis. In Antarctica, there were 21 temperature stations and no precipitation
stations with all but 2 stations being in coastal locations. Like Greenland there was little
interior climatic information on which to base a spline analysis. The climate type at all
the Greenland stations was ET, while in Antarctica they were a mixture of ET and EF.
In order to interpolate the climate type over the whole of Greenland and Antarctica
an alternate method was required. In the case of Greenland, elevation data was used,
with any location <1000 m given a climate type of ET and a location ≥1000 m given a
climate type of EF. In Antarctica the whole continent was set to a climate type of EF,
except around stations that were known to be ET. In Table 1 the criterion that divides
◦
ET from EF is the hottest month being greater than 0 C. Of the Antarctic stations the
◦
hottest month at any station was 1.6 C, so most of the ET stations are borderline EF.
Based on this observation, any location near an ET station with an elevation of <100 m
was set to ET.
4 The world map: discussion and conclusion
20
25
The continental and island maps are combined together to form the world KöppenGeiger climate type map (Fig. 10). Globally the dominant climate class by land area
is arid B (30.2%) followed by cold D (24.6%), tropical A (19.0%), temperate C (13.4%)
and polar E (12.8%). The most common individual climate type by land area is BWh
(14.2%), followed by Aw (11.5%). Of the 30 possible climate types the Csc climate type
never occurs and the Cwc climate type occurs in only 25 pixels (representing 0.002%
of total land area).
In the methodology, a new criterion for locations that satisfy both Cs and Cw (or
458
5
10
15
20
25
Ds and Dw) criteria simultaneously was described and it was applied successfully in
several locations. However, the criteria for the second letter of the C and D climates
was still found to produce inappropriate results, particularly in the high latitudes of
North America, and needs to be further revised. Future revisions based on mutually
exclusive total seasonal precipitation thresholds, similar to the 70% criteria used for the
B climate, may prove more robust than criteria based on the wettest and driest months
of summer and winter.
The updated world Köppen-Geiger climate type map is based on the climatology at
stations over their entire period of record, with each variable individually interpolated
and differs from the recent work of Kottek et al. (2006), which is based on 0.5×0.5 degree gridded temperature and precipitation data for the period 1951 to 2000. Although
broadly similar to the map of Kottek et al. (2006), the present map also deals with
locations that satisfy both Cs and Cw (or Ds and Dw) and has a finer resolution.
The single setting used for the 2-D thin-plate spline with tension was able to successfully interpolate the observed data in the vast majority of cases. For a total of
4279 stations with both precipitation and temperature data, the map differs significantly
at only 3 stations (Kodaikanal, Pune and Aracaju) and in each case a more flexible
spline setting may have improved the map. However, in mountainous regions and in
regions of low station density the addition of elevation into the interpolation procedure
may improve the extrapolation of Köppen-Geiger climate type. In addition to improved
estimates of precipitation and temperate in mountainous regions, elevation information
would limit the influence of isolated high elevation stations extending their climate type
into lowland regions.
The electronic form (in ESRI Arc Grid (raster) and JPG formats) of the world KöppenGeiger climate type map is available at http://www.hydrol-earth-syst-sci.net/????. Also
available at this site are files containing the precipitation and temperature variables for
all stations used in the construction of the map.
Acknowledgements. Australian Research Council Discovery grants DP0449685 and
DP0773016 financially supported this work. F. Peel assisted with the colour scheme
459
used for the maps.
References
5
10
15
20
25
Barnett, T., Zwiers, F., Hegerl, G., Allen, M., Crowley, T., Gillett, N., Hasselmann, K., Jones, P.,
Santer, B., Schnur, R., Stott, P., Taylor, K., and Tett. S.: Detecting and attributing external
influences on the climate system: a review of recent advances, J. Climate, 18, 1291–1314,
2005.
Daly, C.: Guidelines for assessing the suitability of spatial climate data sets, Int. J. Climatol.,
26, 707–721, 2006.
Dick, R. S.: Frequency patterns of arid, semi-arid and humid climates in Queensland, Capricornia, 1, 21–30, 1964.
Essenwanger, O. M.: Classification of climates. In World Survey of Climatology 1C, General
Climatology, Elsevier, Amsterdam, pp. 102, 2001.
Fovell, R. G. and Fovell, M.-Y. C.: Climate zones of the conterminous United States defined
using cluster analysis, J. Climate, 6, 2103–2135, 1993.
Gentilli, J. (Ed.): Climates of Australia and New Zealand, World Survey of Climatology, Vol. 13.
Elsevier, Amsterdam, 405p, 1971.
Gnandesikan, A. and Stouffer, R. J.: Diagnosing atmosphere-ocean general circulation model
errors relevant to the terrestrial biosphere using the Köppen climate classification, Geophys.
Res. Lett., 33, L22701, doi:10.1029/2006GL028098, 2006.
Kalvova, J., Halenka, T., Bezpalcova, K., and Nemesova, I.: Köppen Climate types in observed
and simulated climates, Stud. Geophys. Geod., 47, 185–202, 2003.
Kleidon, A., Fraedrich, K., and Heimann, M.: A green planet versus a desert world: estimating
the maximum effect of vegetation on the land surface climate, Climatic Change, 44, 471–493,
2000.
Köppen, W.: Das geographisca System der Klimate, in: Handbuch der Klimatologie, edited by:
Köppen, W. and Geiger, G., 1. C. Gebr, Borntraeger, 1–44, 1936.
Kottek, M., Grieser, J., Beck, C., Rudolf, B., and Rubel, F.: World map of the Köppen-Geiger
climate classification updated, Meteorol. Zeitschr., 15(3), 259–263, 2006.
Lohmann, U., Sausen, R., Bengtsson, L., Cubasch, U., Perlwitz, J., and Roeckner, E.: The
460
5
10
15
20
25
Köppen climate classification as a diagnostic tool for general circulation models, Clim. Res.,
3, 177–193, 1993.
McMahon, T. A., Finlayson, B. L., Haines, A. T., and Srikanthan, R.: Global Runoff – Continental
Comparisons of Annual Flows and Peak Discharges, Catena Verlag, Cremlingen, 166pp,
1992.
Mitas, L. and Mitasova, H.: General variational approach to the interpolation problem, Comput.
Math. Applic., 16, 983–992, 1998.
Peel, M. C., McMahon, T. A., and Finlayson, B. L.: Continental differences in the variability of
annual runoff – update and reassessment, J. Hydrol., 295, 185–197, 2004.
Peterson, T. C. and Vose, R. S.: An overview of the Global Historical Climatology Network
temperature database, Bull. Am. Meteorol. Soc., 78(12), 2837–2849, 1997.
Russell, R. J.: Dry climates of the United States: I climatic map, University of California, Publications in Geography, 5, 1–41, 1931.
Sanderson, M.: The classification of climates from Pythagoras to Koeppen, Bull. Am. Meteorol.
Soc., 80, 669–673, 1999.
Stern, H., De Hoedt, G., and Ernst, J.: Objective classification of Australian climates Aust.
Meteorol. Mag., 49, 87–96, 2000.
Strahler, A. N.: The Earth Sciences. Harper and Row, New York, 824pp, 1971.
Strahler, A. H. and Strahler, A. N.: Physical Geography : Science and Systems of the Human
Environment. Wiley, New York, 794pp, 2005.
Thornthwaite, C. W.: Problems in the classification of climates, Geogr. Rev., 33(2), 233–255,
1934.
Triantafyllou, G. N. and Tsonis, A. A.: Assessing the ability of the Köppen system to delineate
the general world pattern of climates, Geophys. Res. Lett., 21(25), 2809–2812, 1994.
USGS (2000), HYDRO1k, http://edc.usgs.gov/products/elevation/gtopo30/hydro/ (accessed
16/10/2006).
Wang, M. and Overland, J. E.: Detecting Arctic climate change using Köppen climate classification, Climatic Change, 67, 43–62, 2004.
Wilcock, A. A.: Köppen after fifty years, Ann. Assoc. Am. Geog., 58(1), 12–28, 1968.
461
Table 1. Description of Köppen climate symbols and defining criteria.
1st
2nd
3rd
A
f
m
w
B
W
S
h
k
C
s
w
f
a
b
c
D
s
w
f
a
b
c
d
E
T
F
Description
Criteria*
Tropical
- Rainforest
- Monsoon
- Savannah
Arid
- Desert
- Steppe
- Hot
- Cold
Temperate
- Dry Summer
- Dry Winter
- Without dry season
- Hot Summer
- Warm Summer
- Cold Summer
Cold
- Dry Summer
- Dry Winter
- Without dry season
- Hot Summer
- Warm Summer
- Cold Summer
- Very Cold Winter
Polar
- Tundra
- Frost
Tcold ≥18
Pdry ≥60
Not (Af) & Pdry ≥100–MAP/25
Not (Af) & Pdry <100–MAP/25
MAP<10×Pthreshold
MAP<5×Pthreshold
MAP≥5×Pthreshold
MAT≥18
MAT<18
Thot >10 & 0<Tcold <18
Psdry <40 & Psdry < Pwwet /3
Pwdry <Pswet /10
Not (Cs) or (Cw)
Thot ≥22
Not (a) & Tmon10 ≥4
Not (a or b) & 1≤Tmon10 <4
Thot >10 & Tcold ≤0
Psdry <40 & Psdry <Pwwet /3
Pwdry <Pswet /10
Not (Ds) or (Dw)
Thot ≥22
Not (a) & Tmon10 ≥4
Not (a, b or d)
Not (a or b) & Tcold <–38
Thot <10
Thot > 0
Thot ≤0
*MAP = mean annual precipitation, MAT = mean annual temperature, Thot = temperature of the hottest month, Tcold =
temperature of the coldest month, Tmon10 = number of months where the temperature is above 10, Pdry = precipitation
of the driest month, Psdry = precipitation of the driest month in summer, Pwdry = precipitation of the driest month
in winter, Pswet = precipitation of the wettest month in summer, Pwwet = precipitation of the wettest month in winter,
Pthreshold = varies according to the following rules (if 70% of MAP occurs in winter then Pthreshold = 2 x MAT, if 70%
of MAP occurs in summer then Pthreshold = 2 x MAT + 28, otherwise Pthreshold = 2 x MAT + 14). Summer (winter) is
defined as the warmer (cooler) six month period of ONDJFM and AMJJAS.
462
Table 2. Average monthly and annual precipitation (P) and temperature (T) for Herberton Post
Office, Australia.
P (mm)
◦
T ( C)
J
F
M
A
M
J
J
A
S
O
N
D
Ann
238.4
23.2
229.7
22.8
214.4
21.9
86.0
20.0
46.9
17.9
33.3
16.0
22.0
15.5
18.2
16.5
16.5
18.6
25.3
21.0
77.3
22.4
137.9
23.3
1146
19.9
463
Fig. 1. Location of precipitation stations with 30 or more values for each month.
464
T only
P+T
Fig. 2. Location of temperature stations with 30 or more values for each month. Symbols are
“T only” = temperature data only and “P + T” = both temperature and precipitation data.
465
Fig. 3. Percentage of precipitation and temperature stations with a value for a given month.
466
Af
BWh
Csa
Am
BWk
Csb
Aw
BSh
Cwa
BSk
Cwb
Cfa
Cfb
Fig. 4. Köppen-Geiger climate type map of Africa.
467
Af
BWh
Csa
Cwa
Cfa
Am
BWk
Csb
Cwb
Cfb
Aw
BSh
Cwc
Cfc
BSk
Fig. 5. Köppen-Geiger climate type map of Asia.
468
Dsa
Dwa
Dfa
ET
Dsb
Dwb
Dfb
EF
Dsc
Dwc
Dfc
Dsd
Dwd
Dfd
Af
BWh
Csa
Cwa
Cfa
Am
BWk
Csb
Cwb
Cfb
Aw
BSh
Cfc
BSk
Dsa
Dwa
Dfa
Dsb
Dwb
Dfb
Dsc
Dwc
Dfc
ET
Fig. 6. Köppen-Geiger climate type map of North America.
469
Af
BWh
Am
BWk
Aw
BSh
BSk
Csa
Cwa
Csb
Cwb
Cwc
Cfa
Dsb
Cfb
Dsc
Cfc
Dfb
ET
Dfc
Fig. 7. Köppen-Geiger climate type map of South America.
470
BWh
Csa
Cfa
BWk
Csb
Cfb
Cfc
BSh
BSk
Dsa
Dfa
ET
Dsb
Dfb
EF
Dsc
Dfc
Fig. 8. Köppen-Geiger climate type map of Europe.
471
Af
BWh
Csa
Am
BWk
Csb
Aw
BSh
Cwa
Cfa
Cfb
BSk
Fig. 9. Köppen-Geiger climate type map of Australia.
472
Af
BWh
Csa
Cwa
Cfa
Dsa
Dwa
Dfa
ET
Am
BWk
Csb
Cwb
Cfb
Dsb
Dwb
Dfb
EF
Aw
BSh
Cwc
Cfc
Dsc
Dwc
Dfc
Dsd
Dwd
Dfd
BSk
Fig. 10. Köppen-Geiger climate type map of the World.
473